Training students to prove they're not robots is pushing them to use more AI • Hacker News Board

Re: Training students to prove they're not robots is pushing them to use more AI

#131

Post by lich_king » Sat, Mar 07, 2026, 10:27 PM UTC

PufPufPuf wrote: post #116

lich_king wrote: post #83

Earlier quoted context omitted.

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. The problem isn't that AI detection doesn't work. State of the art…

Eliminating any statistically significant difference between a high-quality human-written text and LLM-written text is exactly what the LLMs are being trained for. At this point, "text is low quality, therefore must be human" is a much stronger signal.

> Eliminating any statistically significant difference between a high-quality human-written text and LLM-written text is exactly what the LLMs are being trained for.

I think you're basing this off a fundamental misunderstanding of what these detectors look for. LLMs generate human-like text, but they also generate roughly the same style and content every time for a given prompt, modulo some small amount of nondeterminism. In essence, they are a very predictable human. Ask Gemini or ChatGPT ten times in a row to write an essay about why AI is awesome, and it will probably strike about the same tone every single time, with similar syntax, idioms, etc.

This is what these tools detect: the default output of "hey ChatGPT, write me a school essay about X". This can be evaded with clever prompting to assume a different writing personality, but there's only so much evasion you can do without making the text weird in other ways.

Re: Training students to prove they're not robots is pushing them to use more AI

#132

Post by theptip » Sat, Mar 07, 2026, 10:29 PM UTC

Someone1234 wrote: post #17

theptip wrote: post #16

This is what terrifies me about the public school system. A revolution has occurred, but it’s unevenly distributed. The schools simply don’t have the flexibility, agility, or frankly it seems motivation to adapt to what has already happened. The ship has sailed; essay writing is no longer a viable form of assessment. The idea to try to build a reliable AI detector is asinine, and fundamentally misunderstands how any…

> This is what terrifies me about the public school system. This has nothing to do with Public School in particular. This is impacting private and university education too.

From what I have seen, (some) private schools are moving faster here; not to say private primary/secondary schools are unaffected, rather that it's worst in public schools.

Re: Training students to prove they're not robots is pushing them to use more AI

#133

Post by oytis » Sat, Mar 07, 2026, 10:31 PM UTC

wongarsu wrote: post #123

lich_king wrote: post #83

Earlier quoted context omitted.

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. The problem isn't that AI detection doesn't work. State of the art…

You can detect if texts from a year ago used AI based on statistical patterns. Nobody is taking issue with that. But once you tell people "we will run these tests to detect if your future submissions are using AI" you create an adversarial environment and your statistical methods will continuously break. Not because statistics is broken, but because you are trying to hit a moving target that doesn't want to be hit. T…

> you create an adversarial environment

Do AI vendors specifically train models to circumvent AI detectors? Why would they?

Re: Training students to prove they're not robots is pushing them to use more AI

#134

Post by armchairhacker » Sat, Mar 07, 2026, 10:32 PM UTC

lich_king wrote: post #83

Zigurd wrote: post #66

The core of the problem the article is about isn't AI or LLMs, it's about scam software that claims to catch cheating. It's crap for the same reasons that crime predictions software is crap. It's selling a panacea, and that kind of product inherently attracts scammers. If your school uses software to detect AI writing, that's a problem with the quality of your school. The people choosing that software are too stupid…

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. The problem isn't that AI detection doesn't work. State of the art…

> When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim.

Different people. I for one have always claimed that fMRI is too coarse-grained for detailed thought detection.

If AI detection "sometimes fails", it doesn't "work". It works well enough to convict someone with other evidence, but when there's no other evidence nor an attempt to get any, it has no good use.

What I propose is simple: grade only closed-book exams, and hold students' phones during the exams. Students don't need 1:1 monitoring, it's the same as 10-20 years ago.

Re: Training students to prove they're not robots is pushing them to use more AI

#135

Post by kayo_20211030 » Sat, Mar 07, 2026, 10:33 PM UTC

thfuran wrote: post #122

kayo_20211030 wrote: post #81

Earlier quoted context omitted.

If you nail the one exam, you get an A+. If you fail it, you get an F. In between, you get what your score says you get.

I understand the proposed grading system but not the reason for selecting that particular system.

It's a crude blade to avoid the issues of AI pollution of weekly submissions, of which few teachers have much confidence that the submission itself was actually written by the student - who's assumed to be learning something.

The OP was about students dumbing down their own work to avoid AI detectors ratting them out. That seems like a big loss.

Re: Training students to prove they're not robots is pushing them to use more AI

#136

Post by lich_king » Sat, Mar 07, 2026, 10:38 PM UTC

wongarsu wrote: post #123

lich_king wrote: post #83

Earlier quoted context omitted.

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. The problem isn't that AI detection doesn't work. State of the art…

You can detect if texts from a year ago used AI based on statistical patterns. Nobody is taking issue with that. But once you tell people "we will run these tests to detect if your future submissions are using AI" you create an adversarial environment and your statistical methods will continuously break. Not because statistics is broken, but because you are trying to hit a moving target that doesn't want to be hit. T…

> You can detect if texts from a year ago used AI based on statistical patterns.

I don't understand your argument. The vendors for these detection tools can acquire recent samples from all frontier models just as easily as you can use them to write essays. There's nothing that requires a one-year delay.

Re: Training students to prove they're not robots is pushing them to use more AI

#137

Post by lich_king » Sat, Mar 07, 2026, 10:44 PM UTC

Zigurd wrote: post #105

lich_king wrote: post #83

Earlier quoted context omitted.

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. The problem isn't that AI detection doesn't work. State of the art…

>> But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading. That's what fortunetellers do. The problem isn't guessing correctly about AI content in writing. The problem is false positives. That's what puts it in the same category is predictive policing scam software. And fortunetelling.

It has nothing to do with predictive policing. I don't understand this example, it has nothing to do with detecting intent. You're looking for evidence of a past misdeed.

False positive and false negative rates are non-zero, as with almost anything, but the tools are pretty good. I encourage you to give them a try. Pangram is a good state-of-the-art choice and you can try it for free. They also publish evals and other data about their approach.

Re: Training students to prove they're not robots is pushing them to use more AI

#138

Post by Paracompact » Sat, Mar 07, 2026, 10:44 PM UTC

Onavo wrote: post #15

Paracompact wrote: post #2

Grade school has never been kind to genuine writers. It reminds me of SAT essays that favored formulaic writing, because guess what: the grading criteria were formulaic! I think grading in general can be stymying for students' motivation and creative drives.

Being able to write in a formulaic manner is a skill though. Not being able to write properly is not great for communication.

There was nothing useful about the particular formula they were teaching. It wouldn't even be useful for a bureaucrat. It only tested how well you knew the formula, how confidently you simplified inherently nuanced topics, and how lucky you got that the random underpaid SAT grader (usually a teacher looking for a pittance of extra cash) thought your essay fit the rubrics they were given.

Good riddance to the thing.

Re: Training students to prove they're not robots is pushing them to use more AI

#139

Post by BeetleB » Sat, Mar 07, 2026, 10:48 PM UTC

armchairhacker wrote: post #109

kayo_20211030 wrote: post #40

Perhaps we should not grade students on weekly, or other occasional, writing during the term or semester. How about going back to the old system where, apart from experimental lab work, nothing is graded until the end of the term? All weekly assignments should just be considered prep for one exam at the end of the term where the student has an opportunity to demonstrate mastery of the course's subject matter. They ca…

Why not multiple exams? In fact, why not many exams? Sure, it requires more resources, but it shouldn't require much more: - We've had multiple exams before AI, and I don't see how AI makes it any harder. Obviously these are closed-book - Schools should already be banning phones in class (and colleges have insane tuitions, they can afford more exams) - The students who go out of their way to cheat - as long as they'r…

Indeed. Many of my technical undergrad courses were very exam heavy. Typically 3-5 midterms and one final. Sometimes the final was as little as 10% of the grade. The idea was that if you'd done well throughout the semester you can relax during finals weeks.

Homework was assigned but not graded.

Periodic tests is the way to go.

I hated courses where the final was more than 30%. Forget 100%.

Re: Training students to prove they're not robots is pushing them to use more AI

#140

Post by csa » Sat, Mar 07, 2026, 10:54 PM UTC

kayo_20211030 wrote: post #60

eudamoniac wrote: post #55

Earlier quoted context omitted.

Schools stopped doing that because students largely refuse to prepare. Testing throughout the year is like a CI pipeline and is shown to work better for the median student.

Students are neither generally stupid nor constitutionally lazy. I sense that when expectations are clear they'll often surprise you with diligence. We should trust them to do the right thing. If they do, it's an A; and if not, it's less than that.

> I sense that when expectations are clear they'll often surprise you with diligence.

Data does not support your sense.

Most students do not have good time management skills, usually because they have no models and/or have not been taught these skills.

Furthermore, continuous feedback, whether graded or not, has been found to be more effective than one-shot feedback.

Evaluation and assessment is a complex topic towards which many people (not necessarily you) want to take an overly simplified approach.

There are trade offs for any system that is chosen. The organizations providing the grades have to decide what their priorities are (e.g., time, accuracy, etc.).